5
$\begingroup$

In definition 1.1 of these notes on factorization systems, (III) calls the factorization functorial if given the solid diagram described, there's a unique horizontal arrow making both squares commute. Why does this deserve the name 'functorial'?

$\endgroup$
6
$\begingroup$

Let $[n]$ denote the category $0\to 1\to\cdots\to n$. Given a factorization system $(L, R)$ in a category $\mathcal{C}$, we can choose a map $F_0\colon\operatorname{Ob}\mathcal{C}^{[1]}\to\operatorname{Ob}\mathcal{C}^{[2]}$ which maps an object $x\xrightarrow{f} y$ in $\operatorname{Ob}\mathcal{C}^{[1]}$ to an object $x\xrightarrow{e} E\xrightarrow{m} y$ in $\mathcal{C}^{[2]}$ such that $e\in L$, $m\in R$ and $f=me$.

If your factorization system is functorial, this extends to a functor $F\colon\mathcal{C}^{[1]}\to\mathcal{C}^{[2]}$ mapping a morphism $(u,v)$ in $\mathcal{C}^{[1]}$ to a morphism $(u,\omega,v)$ in $\mathcal{C}^{[2]}$ where $\omega$ is the unique morphism required in (III), making the respective diagram commutative. The uniqueness guaranties that $F$ is indeed a functor, i.e. that given two morphism $(u,v),(u',v')\in\mathcal{C}^{[1]}$, we have

$$ F((u,v))\circ F(u',v') = (u,\omega,v)\circ (u',\omega',v') = (u\circ v',\omega\circ\omega',v\circ v') = F((u\circ u',v\circ v'))$$

Also note that even though the choice of $F_0$ is not unique, (III) guarantees that it is unique up to isomorphisms. To prove this, take two factorizations $(e, m)$ and $(e',m')$ of the same morphism $f$ and consider the diagrams $$ \require{AMScd} \begin{CD} x@>\operatorname{id}>> x @>\operatorname{id}>> x\\ @V{e}VV @V{e'}VV @V{e}VV \\ E@>\omega>>E' @>\omega'>> E \\ @V{m}VV @V{m'}VV @V{m}VV\\ y@>\operatorname{id}>> y @>\operatorname{id}>> y \end{CD} $$ and $$ \begin{CD} x@>\operatorname{id}>> x\\ @V{e}VV @V{e}VV \\ E@>\omega'\circ\omega>> E \\ @V{m}VV @V{m}VV\\ y@>\operatorname{id}>> y \end{CD} $$ By (III): $\omega'\circ\omega$ is unique, hence $\omega'\circ\omega = \operatorname{id}$. Thus our factorization functor $F$ is determined up to isomorphisms, and it makes sense to talk about the associated factorization functor.

$\endgroup$
  • $\begingroup$ Quick question - in the last sentence you say $F$ is determined up to isomorphism, but it seems $F$ is only determined up to isomorphism object-wise, i.e $F(u,v)$ is determined up to iso for each object $(u,v)$. Why is it true that all extensions are naturally isomorphic as functors? $\endgroup$ – Arrow Apr 26 '16 at 15:39
  • $\begingroup$ @Arrow I thin you mixed something up here. $F_0$ is uniquely determined up to isomorphisms and only defined on objects $x\xrightarrow{f} y$ in $\mathcal{C}^{[1]}$. The extension to a functor $F$ is then given by mapping morphisms $(u,v)$ in $\mathcal{C}^{[1]}$ to morphisms $(u,\omega, v)$. I'm not sure right now if this extension is unique, or if you could do something weird like sending all morphisms in $\mathbf{Grp}^{[1]}$ to those mapping everything to the neutral element. But it's at least the unique natural extension. Where uniqueness is guaranteed by (III). $\endgroup$ – roman Apr 28 '16 at 8:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.